An experimental evaluation of novelty detection methods

Xuemei Ding, Yuhua Li, Ammar Belatreche, LP Maguire

Research output: Contribution to journalArticle

61 Citations (Scopus)

Abstract

Novelty detection is especially important for monitoring safety-critical systems in which novel conditions rarely occur and knowledge about novelty in that system is often limited or unavailable. There are a large number of studies in the area of novelty detection, but there is a lack of a comprehensive experimental evaluation of existing novelty detection methods. This paper aims to fill this void by conducting experimental evaluation of representative novelty detection methods. It presents a state-of-the-art review of novelty detection, with a focus on methods reported in the last few years. In addition, a rigorous comparative evaluation of four widely used methods, representative of different categories of novelty detectors, is carried out using 10 benchmark datasets with different scale, dimensionality and problem complexity. The experimental results demonstrate that the k-NN novelty detection method exhibits competitive overall performance to the other methods in terms of the AUC metric.
LanguageEnglish
Pages313-327
JournalNeurocomputing
Volume135
DOIs
Publication statusPublished - 5 Jul 2014

Fingerprint

Detectors
Monitoring

Cite this

Ding, Xuemei ; Li, Yuhua ; Belatreche, Ammar ; Maguire, LP. / An experimental evaluation of novelty detection methods. In: Neurocomputing. 2014 ; Vol. 135. pp. 313-327.
@article{a0b808427f264b60bb24e7e6ca62fb5c,
title = "An experimental evaluation of novelty detection methods",
abstract = "Novelty detection is especially important for monitoring safety-critical systems in which novel conditions rarely occur and knowledge about novelty in that system is often limited or unavailable. There are a large number of studies in the area of novelty detection, but there is a lack of a comprehensive experimental evaluation of existing novelty detection methods. This paper aims to fill this void by conducting experimental evaluation of representative novelty detection methods. It presents a state-of-the-art review of novelty detection, with a focus on methods reported in the last few years. In addition, a rigorous comparative evaluation of four widely used methods, representative of different categories of novelty detectors, is carried out using 10 benchmark datasets with different scale, dimensionality and problem complexity. The experimental results demonstrate that the k-NN novelty detection method exhibits competitive overall performance to the other methods in terms of the AUC metric.",
author = "Xuemei Ding and Yuhua Li and Ammar Belatreche and LP Maguire",
year = "2014",
month = "7",
day = "5",
doi = "10.1016/j.neucom.2013.12.002",
language = "English",
volume = "135",
pages = "313--327",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",

}

An experimental evaluation of novelty detection methods. / Ding, Xuemei; Li, Yuhua; Belatreche, Ammar; Maguire, LP.

In: Neurocomputing, Vol. 135, 05.07.2014, p. 313-327.

Research output: Contribution to journalArticle

TY - JOUR

T1 - An experimental evaluation of novelty detection methods

AU - Ding, Xuemei

AU - Li, Yuhua

AU - Belatreche, Ammar

AU - Maguire, LP

PY - 2014/7/5

Y1 - 2014/7/5

N2 - Novelty detection is especially important for monitoring safety-critical systems in which novel conditions rarely occur and knowledge about novelty in that system is often limited or unavailable. There are a large number of studies in the area of novelty detection, but there is a lack of a comprehensive experimental evaluation of existing novelty detection methods. This paper aims to fill this void by conducting experimental evaluation of representative novelty detection methods. It presents a state-of-the-art review of novelty detection, with a focus on methods reported in the last few years. In addition, a rigorous comparative evaluation of four widely used methods, representative of different categories of novelty detectors, is carried out using 10 benchmark datasets with different scale, dimensionality and problem complexity. The experimental results demonstrate that the k-NN novelty detection method exhibits competitive overall performance to the other methods in terms of the AUC metric.

AB - Novelty detection is especially important for monitoring safety-critical systems in which novel conditions rarely occur and knowledge about novelty in that system is often limited or unavailable. There are a large number of studies in the area of novelty detection, but there is a lack of a comprehensive experimental evaluation of existing novelty detection methods. This paper aims to fill this void by conducting experimental evaluation of representative novelty detection methods. It presents a state-of-the-art review of novelty detection, with a focus on methods reported in the last few years. In addition, a rigorous comparative evaluation of four widely used methods, representative of different categories of novelty detectors, is carried out using 10 benchmark datasets with different scale, dimensionality and problem complexity. The experimental results demonstrate that the k-NN novelty detection method exhibits competitive overall performance to the other methods in terms of the AUC metric.

U2 - 10.1016/j.neucom.2013.12.002

DO - 10.1016/j.neucom.2013.12.002

M3 - Article

VL - 135

SP - 313

EP - 327

JO - Neurocomputing

T2 - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

ER -